SAR Image Despeckling Based on Local Noise Variance
Shaona Wang, Zhenghua Zhang, Di Wang · 2023
Synthetic Aperture Radar (SAR) image despeckling is a basic task in SAR image processing. The size of noise variance affects the level of image speckle removal. Excessive variance can lead to the loss of details in the image, while conversely, it can still result in low-frequency noise in the denoised image. In this paper, we start from a sample matrix consisting of similar patches and calculate the local noise variance of each sample matrix. The PCA coefficients are contracted using the local noise variance with LMMSE. For ease of computation, we degenerate the coefficient contraction method to a kernel-paradigm minimization form using the mean of the local noise variance. Finally, the estimation bias induced by linear versus nonlinear signals is balanced by dividing the patches into homogeneous and heterogeneous categories using a region partitioning technique. The despeckling experimental results of simulated and real SAR images show good performance in both quantitative measurement and visual perception.